Look-ahead bias

The most common version is embarrassingly simple: a signal computed on the day's close but acted on at that same day's open, as if the close had already happened. Rank the day's winners after the fact and trade them from the morning bar, and the backtest is trading tomorrow's newspaper. Nobody wrote that rule on purpose. It slips in through a spreadsheet formula, a lookback window off by one bar, an indicator with a hidden peek.

Repainting indicators are the sneakier version. The indicator quietly revises its own past values as new bars arrive, so the signal on your chart right now never actually fired that cleanly in real time. Backtest against the repainted history and every entry looks prescient. It is prescient. It was computed with knowledge of what came next.

This is what makes look-ahead the most dangerous bug in backtesting: it produces the most convincing fake in the whole audit. A strategy with no real edge usually shows it — the equity curve is choppy, thin, unconvincing. A strategy with look-ahead bias does the opposite. It produces a smooth, high-Sharpe curve that could never have been traded, because a chunk of its trades were decided with information from the future. That's the leak. It passes every surface check a tired reviewer runs.

Look-ahead is a close cousin of the intrabar fill assumption — both let a strategy touch a price it couldn't have reached live. The intrabar version is a sequencing mistake inside a single candle. Look-ahead is a straight leak across time, and it's the sharper of the two.

In our pipeline, every strategy that clears the numbers gets a second pass built specifically to catch this. Our strongest model adversarially audits each survivor, hunting for a fill that peeked or a value that leaked from a bar that hadn't closed yet, and we cross-check timing against walk-forward splits instead of one in-sample run.

It's a smaller slice of our rejects than no real edge, which alone accounts for roughly half of the strategies we reject. But it's the costlier failure. These are the strategies that look best right up until you trade them for real, and then they fall apart.